How to Continue When 25 Tool Call Limit is Reached

When Cursor notifies me with:

“Note: we default stop the agent after 25 tool calls. Please ask the agent to continue manually.”

What should I say to resume the task? Would “continue this task” work, or is there a better command?

Also, I’d love to see an option to set a custom limit or even allow unlimited tool calls. Is this something that could be added in the future?

Old Cursor version (pre 0.46) had. this. you can write: continue.
Newer cursor versions 0.46+ have an option right under the chat output to resume.

Both options cost a request (same amount as the initial request, since you continue with detailed steps). The actual request count will depend on the model you used. see Models in Cursor Docu.

Can I exceed this limit (25 tool calls)?

No, as the docu shows and the Cursor IDE tells you to. You can continue/resume the process.

how to let cursor automatically continue ?

2 Likes

you click on “Resume” in the chat or if your version is older and does not have that, then you can type new message ‘continue’ or similar.

i mean automatically , like an option ,or api that i can set to resume or continue to automat continue ~

some task may last long time and many action,it often reach the limitation

3 Likes

I guess we would need to build an other cursor on top of cursor to press that button automatically? ..

And then someone sets that setting and complains that cursor charged and never stopped :slight_smile:

But yes some kind of Auto-run Resume setting (with a clear max repeat count) would be reasonable for advanced users.

The completion sound helps actually a lot if one doesnt want to keep watching Cursors activity in that window.

See also

2 Likes

can you just add a switch in full-auto-run mode in feature section of Cursor setting so that if this is checked it will automatically continue forever? it is really annoying…

1 Like

I wish they had something like that…I am having the same problem here

Not quite sure who on the Cursor team marked the topic as resolved. I am still running into this issue. I would like to have a setting that automatically bypass the 25 tool call limit so I do not have to babysit Cursor by pressing resume. This is a bad Dev Experience to have to do that. Is there a timeline to automatically configure a set limit instead of the arbitrary 25 tool calls limit?

@trungbui2000 Hi and welcome to the forum.

The question of “How to continue…” was answered.

There is a separate feature request for increasing the tool call limit.

Currently MAX mode supports 200 tool calls, as the larger size of the MAX mode context allows more information to be passed in the call to AI.

The tool call limit is not related to the context size, GitHub Copilot allows an arbitrary limit configured by the user. You make it sound like it’s related which is false.

The tool calls are not aggregated or anything like that it’s just an arbitrary limit set for the number of tool calls allowed in sequence without manual interruption.

Hey, you’re right that other tools may have different config but what happens in background is not visible to user.

Tool calls do affect context, in size and may affect accuracy of AI responses as they fetch or send information. So its not like there is no impact. At some point, accumulated tool call data can actually confuse the AI with information thats not relevant to the current task.

The API providers (Anthropic, Gemini, OpenAI) typically allow one tool call per API request. Cursor’s orchestration combines multiple calls, and each tool call, third-party or internal, requires preparation, processing, and response handling that factors into the overall request cost. So both context and cost have an affect.

I would like to see configurable limits too, thats why I also linked the feature request! The more people request this the more will this be noticeable.

The OP’s original question about how to continue has been answered.

Hey!

Dan Perks from Cursor commented this

With no mention of context. I mean a tool call will add to context but it doesn’t add more if you have more of them executed in sequence than if you would do one at a time. If it would take another request just to deal with some internal limitation people would be fine with that.

Yeah it’s good if we can drive everyone to that one Feature request, sadly it looks like it’s hidden from the search and not shown under total votes.

Yeah seems like the original question is answered

And this gives some hope